DOCSLIB.ORG

1-Completions of a Poset

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
1-Completions of a Poset Order (2013) 30:39–64 DOI 10.1007/s11083-011-9226-0 1-completions of a Poset Mai Gehrke · Ramon Jansana · Alessandra Palmigiano Received: 10 October 2010 / Accepted: 5 July 2011 / Published online: 27 July 2011 © The Author(s) 2011. This article is published with open access at Springerlink.com Abstract A join-completion of a poset is a completion for which each element is obtainable as a supremum, or join, of elements from the original poset. It is well known that the join-completions of a poset are in one-to-one correspondence with the closure systems on the lattice of up-sets of the poset. A 1-completion of a poset is a completion for which, simultaneously, each element is obtainable as a join of meets of elements of the original poset and as a meet of joins of elements from the original poset. We show that 1-completions are in one-to-one correspondence with certain triples consisting of a closure system of down-sets of the poset, a closure system of up-sets of the poset, and a binary relation between these two systems. Certain 1-completions, which we call compact, may be described just by a collection of filters and a collection of ideals, taken as parameters. The compact 1-completions of a poset include its MacNeille completion and all its join- and all The research of the second author has been partially supported by SGR2005-00083 research grant of the research funding agency AGAUR of the Generalitat de Catalunya and by the MTM2008-01139 research grant of the Spanish Ministry of Education and Science. The research of the third author has been supported by the VENI grant 639.031.726 of the Netherlands Organisation for Scientific Research (NWO). M. Gehrke (B) FNWI, IMAPP, Radboud Universiteit Nijmegen, P.O. Box 9010, 6500 GL Nijmegen, The Netherlands e-mail: [email protected] R. Jansana Dept. Lògica, Història i Filosofia, de la Ciència, Universitat de Barcelona, Montalegre, 6, 08001, Barcelona, Spain A. Palmigiano FNWI, ILLC, Universiteit van Amsterdam, P.O. Box 94242, 1090 GE Amsterdam, The Netherlands 40 Order (2013) 30:39–64 its meet-completions. These completions also include the canonical extension of the given poset, a completion that encodes the topological dual of the poset when it has one. Finally, we use our parametric description of 1-completions to compare the canonical extension to other compact 1-completions identifying its relative merits. Keywords Completions of a poset · Canonical extensions 1 Introduction Embedding partially ordered sets in complete lattices is central to the solution of many problems. This is most famously exemplified by the construction of the real numbers from the rational numbers [6]. Dedekind’s construction, generalised to arbitrary posets by MacNeille [24], yields a completion for any poset P.The MacNeille completion of a poset is minimal in some sense but it does not have good algebraic preservation properties, e.g. it does not preserve homomorphisms, equational properties, and the like. For this purpose one would rather want a greatest completion, or in other words, a free completion. One-sided free completions exist and have indeed been used to great success, e.g. in domain theory. However, two- sided free completions (that is, completions that are free with respect to complete lattice homomorphisms rather than just with respect to complete join-morphisms or complete meet-morphisms) are well known not to exist—see e.g. [20, Section 4.7]. In [21, 22] Bjarni Jónsson and Alfred Tarski introduced a two-sided completion for Boolean algebras, called canonical extension, which is the free completion within the class of completely distributive (or equivalently, atomic) complete Boolean algebras. The canonical extension is closely related to duality theory and provides an algebraic approach to topological duality [15, 16]. As a completion, it has the virtue of preserving more equational properties than the MacNeille completion [13] and it has played a substantial role, in a more or less hidden way, in the semantic study of many logics including modal and intuitionistic logic [4, 7]. More recently, canonical extension has been generalised to encompass distributive lattices [14], lattices [12], and even posets [8] and the theory has been applied in logic and algebra, e.g. [1, 8, 15, 16]. Canonical extensions have the drawback that they do not preserve existing infinite meets and joins and, in the other direction, they do not add finite joins and meets in a free way. Depending on the application, e.g. fixed- point logics versus very weak fragments of logic [18], different completions may be appropriate. In this paper we study a class of two-sided completions encompassing both the MacNeille completion and the canonical extension. The identifying property of these completions is their placement in a meet-join complexity hierarchy for completions. This hierarchy was first brought to the attention of the first author by Keith Kearnes in a private communication and was subsequently studied in relation to canonical extension in [17]. The completions that we focus on here are called 1-completions and figure in the hierarchy as the completions for which each element is reachable by joins of meets and by meets of joins of elements from the original poset. Included among the 1-completions are the 0-completions which are the well-known join- completions as well as the 0-completions which are the meet-completions. The canonical extension of a poset is an example of a 1-completion which, in general, is Order (2013) 30:39–64 41 neither a 0-completion nor a 0-completion. Any 1-completion can be built from a collection of up-sets and a collection of down-sets of the original poset. However, possibly somewhat surprisingly, specifying collections of up-sets and down-sets is not in general sufficient for specifying a 1-completion. Our main result (Theorem 3.3) is a complete classification of the 1-completions of a poset P in terms of certain polarities (F, I, R) where F is a closure system of up-sets of P and I is a closure system of down-sets of P and R is a relation from F to I satisfying four simple conditions. The relation essentially specifies which meets of down-sets are below which joins of up-sets in the completion. Given a poset P, and closure systems F and I of up-sets and down-sets of P, respectively, there is always a smallest relation R from F to I satisfying the four conditions of Theorem 3.3. It is the relation of non-empty intersection, that is, for every F ∈ F and every I ∈ I, FRI iff F ∩ I = ∅. (1) In fact, even for collections of up-sets and down-sets of P which are not closure systems, one may consider the polarity obtained by this relation of non-empty intersection and thus reach a larger class of 1-completions using only this one relation. This is the idea behind what we call compact 1-completions and leads to notions of compactness and denseness which are parametric in collections F and I of up-sets and down-sets, respectively. We show that, for each choice of F and I containing the principal up-sets and the principal down-sets, respectively, there is (up to isomorphism) a unique completion of P which is (F, I)-compact and (F, I)- dense. Thus parametric compactness and denseness allow a study of 1-completions in which the relational component may be omitted in the sense that it is fixed to be the relation of non-intersection as given by Eq. 1. The class of parametrically compact and dense completions includes well-known completions such as the MacNeille completion, the canonical extension, and all join- and meet-completions of a given poset. However, we provide examples of posets for which it is not possible to describe all 1-completions as compact and dense with respect to some F and some I. Thus the relational component of our classification result is in general necessary. With the purpose of comparing canonical extension as defined in [8] to other completions which are (F, I)-compact and (F, I)-dense for some choice of F and I, we give conditions on F and I corresponding to various properties of the (F, I)-compact and (F, I)-dense completion of P, concentrating on properties which are central to applications in logic. The paper is organised as follows. In Section 2 we introduce 1-completions and show that MacNeille completions and Galois connections are central to the study of these completions. In Section 3 we briefly review the necessary preliminaries on Galois connections including the abstract characterisation of the lattice of Galois closed sets of a polarity as well as its construction as the MacNeille completion of a certain ‘intermediate’ structure. These tools then allow us to give a complete classification of all 1-completions of a given poset. In Section 4 we introduce what we call 1-polarities which are based on collections of up-sets and down-sets that are not necessarily closure systems. In Section 5 we study parametric compactness and in Section 6 we identify the properties required of F and I in order that the corresponding completion have various desirable properties. 42 Order (2013) 30:39–64 2 1-completions We fix a poset P. A completion of P is an embedding of P inacompletelattice.Tobe precise, an order embedding is a map e: P → Q from a poset P into a poset Q so that, for every x, y ∈ P, we have x ≤ y if and only if e(x) ≤ e(y).Aposet extension of P is a pair (e, Q),whereQ is a partially ordered set and e: P → Q is an order embedding.
Load more

Recommended publications
  • On Scattered Convex Geometries
    ON SCATTERED CONVEX GEOMETRIES KIRA ADARICHEVA AND MAURICE POUZET Abstract. A convex geometry is a closure space satisfying the anti-exchange axiom. For several types of algebraic convex geometries we describe when the collection of closed sets is order scat- tered, in terms of obstructions to the semilattice of compact elements. In particular, a semilattice Ω(η), that does not appear among minimal obstructions to order-scattered algebraic modular lattices, plays a prominent role in convex geometries case. The connection to topological scat- teredness is established in convex geometries of relatively convex sets. 1. Introduction We call a pair X; φ of a non-empty set X and a closure operator φ 2X 2X on X a convex geometry[6], if it is a zero-closed space (i.e. ) and φ satisfies the anti-exchange axiom: ( ) ∶ → x A y and x∅ =A∅imply that y A x for all x y in X and all closed A X: ∈ ∪ { } ∉ ∉ ∪ { } The study of convex geometries in finite≠ case was inspired by their⊆ frequent appearance in mod- eling various discrete structures, as well as by their juxtaposition to matroids, see [20, 21]. More recently, there was a number of publications, see, for example, [4, 43, 44, 45, 48, 7] brought up by studies in infinite convex geometries. A convex geometry is called algebraic, if the closure operator φ is finitary. Most of interesting infinite convex geometries are algebraic, such as convex geometries of relatively convex sets, sub- semilattices of a semilattice, suborders of a partial order or convex subsets of a partially ordered set.
    [Show full text]
  • Advanced Discrete Mathematics Mm-504 &
    1 ADVANCED DISCRETE MATHEMATICS M.A./M.Sc. Mathematics (Final) MM-504 & 505 (Option-P3) Directorate of Distance Education Maharshi Dayanand University ROHTAK – 124 001 2 Copyright © 2004, Maharshi Dayanand University, ROHTAK All Rights Reserved. No part of this publication may be reproduced or stored in a retrieval system or transmitted in any form or by any means; electronic, mechanical, photocopying, recording or otherwise, without the written permission of the copyright holder. Maharshi Dayanand University ROHTAK – 124 001 Developed & Produced by EXCEL BOOKS PVT. LTD., A-45 Naraina, Phase 1, New Delhi-110 028 3 Contents UNIT 1: Logic, Semigroups & Monoids and Lattices 5 Part A: Logic Part B: Semigroups & Monoids Part C: Lattices UNIT 2: Boolean Algebra 84 UNIT 3: Graph Theory 119 UNIT 4: Computability Theory 202 UNIT 5: Languages and Grammars 231 4 M.A./M.Sc. Mathematics (Final) ADVANCED DISCRETE MATHEMATICS MM- 504 & 505 (P3) Max. Marks : 100 Time : 3 Hours Note: Question paper will consist of three sections. Section I consisting of one question with ten parts covering whole of the syllabus of 2 marks each shall be compulsory. From Section II, 10 questions to be set selecting two questions from each unit. The candidate will be required to attempt any seven questions each of five marks. Section III, five questions to be set, one from each unit. The candidate will be required to attempt any three questions each of fifteen marks. Unit I Formal Logic: Statement, Symbolic representation, totologies, quantifiers, pradicates and validity, propositional logic. Semigroups and Monoids: Definitions and examples of semigroups and monoids (including those pertaining to concentration operations).
    [Show full text]
  • Simple Laws About Nonprominent Properties of Binary Relations
    Simple Laws about Nonprominent Properties of Binary Relations Jochen Burghardt jochen.burghardt alumni.tu-berlin.de Nov 2018 Abstract We checked each binary relation on a 5-element set for a given set of properties, including usual ones like asymmetry and less known ones like Euclideanness. Using a poor man's Quine-McCluskey algorithm, we computed prime implicants of non-occurring property combinations, like \not irreflexive, but asymmetric". We considered the laws obtained this way, and manually proved them true for binary relations on arbitrary sets, thus contributing to the encyclopedic knowledge about less known properties. Keywords: Binary relation; Quine-McCluskey algorithm; Hypotheses generation arXiv:1806.05036v2 [math.LO] 20 Nov 2018 Contents 1 Introduction 4 2 Definitions 8 3 Reported law suggestions 10 4 Formal proofs of property laws 21 4.1 Co-reflexivity . 21 4.2 Reflexivity . 23 4.3 Irreflexivity . 24 4.4 Asymmetry . 24 4.5 Symmetry . 25 4.6 Quasi-transitivity . 26 4.7 Anti-transitivity . 28 4.8 Incomparability-transitivity . 28 4.9 Euclideanness . 33 4.10 Density . 38 4.11 Connex and semi-connex relations . 39 4.12 Seriality . 40 4.13 Uniqueness . 42 4.14 Semi-order property 1 . 43 4.15 Semi-order property 2 . 45 5 Examples 48 6 Implementation issues 62 6.1 Improved relation enumeration . 62 6.2 Quine-McCluskey implementation . 64 6.3 On finding \nice" laws . 66 7 References 69 List of Figures 1 Source code for transitivity check . .5 2 Source code to search for right Euclidean non-transitive relations . .5 3 Timing vs. universe cardinality .
    [Show full text]
  • Completely Representable Lattices
    Completely representable lattices Robert Egrot and Robin Hirsch Abstract It is known that a lattice is representable as a ring of sets iff the lattice is distributive. CRL is the class of bounded distributive lattices (DLs) which have representations preserving arbitrary joins and meets. jCRL is the class of DLs which have representations preserving arbitrary joins, mCRL is the class of DLs which have representations preserving arbitrary meets, and biCRL is defined to be jCRL ∩ mCRL. We prove CRL ⊂ biCRL = mCRL ∩ jCRL ⊂ mCRL =6 jCRL ⊂ DL where the marked inclusions are proper. Let L be a DL. Then L ∈ mCRL iff L has a distinguishing set of complete, prime filters. Similarly, L ∈ jCRL iff L has a distinguishing set of completely prime filters, and L ∈ CRL iff L has a distinguishing set of complete, completely prime filters. Each of the classes above is shown to be pseudo-elementary hence closed under ultraproducts. The class CRL is not closed under elementary equivalence, hence it is not elementary. 1 Introduction An atomic representation h of a Boolean algebra B is a representation h: B → ℘(X) (some set X) where h(1) = {h(a): a is an atom of B}. It is known that a representation of a Boolean algebraS is a complete representation (in the sense of a complete embedding into a field of sets) if and only if it is an atomic repre- sentation and hence that the class of completely representable Boolean algebras is precisely the class of atomic Boolean algebras, and hence is elementary [6]. arXiv:1201.2331v3 [math.RA] 30 Aug 2016 This result is not obvious as the usual definition of a complete representation is thoroughly second order.
    [Show full text]
  • On Birkhoff's Common Abstraction Problem
    F. Paoli On Birkho®'s Common C. Tsinakis Abstraction Problem Abstract. In his milestone textbook Lattice Theory, Garrett Birkho® challenged his readers to develop a \common abstraction" that includes Boolean algebras and lattice- ordered groups as special cases. In this paper, after reviewing the past attempts to solve the problem, we provide our own answer by selecting as common generalization of BA and LG their join BA_LG in the lattice of subvarieties of FL (the variety of FL-algebras); we argue that such a solution is optimal under several respects and we give an explicit equational basis for BA_LG relative to FL. Finally, we prove a Holland-type representation theorem for a variety of FL-algebras containing BA _ LG. Keywords: Residuated lattice, FL-algebra, Substructural logics, Boolean algebra, Lattice- ordered group, Birkho®'s problem, History of 20th C. algebra. 1. Introduction In his milestone textbook Lattice Theory [2, Problem 108], Garrett Birkho® challenged his readers by suggesting the following project: Develop a common abstraction that includes Boolean algebras (rings) and lattice ordered groups as special cases. Over the subsequent decades, several mathematicians tried their hands at Birkho®'s intriguing problem. Its very formulation, in fact, intrinsically seems to call for reiterated attempts: unlike most problems contained in the book, for which it is manifest what would count as a correct solution, this one is stated in su±ciently vague terms as to leave it open to debate whether any proposed answer is really adequate. It appears to us that Rama Rao puts things right when he remarks [28, p.
    [Show full text]
  • Steps in the Representation of Concept Lattices and Median Graphs Alain Gély, Miguel Couceiro, Laurent Miclet, Amedeo Napoli
    Steps in the Representation of Concept Lattices and Median Graphs Alain Gély, Miguel Couceiro, Laurent Miclet, Amedeo Napoli To cite this version: Alain Gély, Miguel Couceiro, Laurent Miclet, Amedeo Napoli. Steps in the Representation of Concept Lattices and Median Graphs. CLA 2020 - 15th International Conference on Concept Lattices and Their Applications, Sadok Ben Yahia; Francisco José Valverde Albacete; Martin Trnecka, Jun 2020, Tallinn, Estonia. pp.1-11. hal-02912312 HAL Id: hal-02912312 https://hal.inria.fr/hal-02912312 Submitted on 5 Aug 2020 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. Steps in the Representation of Concept Lattices and Median Graphs Alain Gély1, Miguel Couceiro2, Laurent Miclet3, and Amedeo Napoli2 1 Université de Lorraine, CNRS, LORIA, F-57000 Metz, France 2 Université de Lorraine, CNRS, Inria, LORIA, F-54000 Nancy, France 3 Univ Rennes, CNRS, IRISA, Rue de Kérampont, 22300 Lannion, France {alain.gely,miguel.couceiro,amedeo.napoli}@loria.fr Abstract. Median semilattices have been shown to be useful for deal- ing with phylogenetic classication problems since they subsume me- dian graphs, distributive lattices as well as other tree based classica- tion structures. Median semilattices can be thought of as distributive _-semilattices that satisfy the following property (TRI): for every triple x; y; z, if x ^ y, y ^ z and x ^ z exist, then x ^ y ^ z also exists.
    [Show full text]
  • The Complexity of Embedding Orders Into Small Products of Chains Olivier Raynaud, Eric Thierry
    The complexity of embedding orders into small products of chains Olivier Raynaud, Eric Thierry To cite this version: Olivier Raynaud, Eric Thierry. The complexity of embedding orders into small products of chains. 2008. hal-00261826 HAL Id: hal-00261826 https://hal.archives-ouvertes.fr/hal-00261826 Preprint submitted on 10 Mar 2008 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. THE COMPLEXITY OF EMBEDDING ORDERS INTO SMALL PRODUCTS OF CHAINS. 1 2 O. RAYNAUD AND E. THIERRY 1 LIMOS, Universit´e Blaise Pascal, Campus des C´ezeaux, Clermont-Ferrand, France. E-mail address: [email protected] 2 LIAFA, Universit´e Paris 7 & LIP, ENS Lyon, France. E-mail address: [email protected] Abstract. Embedding a partially ordered set into a product of chains is a classical way to encode it. Such encodings have been used in various fields such as object oriented programming or distributed computing. The embedding associates with each element a sequence of integers which is used to perform comparisons between elements. A critical measure is the space required by the encoding, and several authors have investigated ways to minimize it, which comes to embedding partially ordered sets into small products of chains.
    [Show full text]
  • Partial Orders — Basics
    Partial Orders — Basics Edward A. Lee UC Berkeley — EECS EECS 219D — Concurrent Models of Computation Last updated: January 23, 2014 Outline Sets Join (Least Upper Bound) Relations and Functions Meet (Greatest Lower Bound) Notation Example of Join and Meet Directed Sets, Bottom Partial Order What is Order? Complete Partial Order Strict Partial Order Complete Partial Order Chains and Total Orders Alternative Definition Quiz Example Partial Orders — Basics Sets Frequently used sets: • B = {0, 1}, the set of binary digits. • T = {false, true}, the set of truth values. • N = {0, 1, 2, ···}, the set of natural numbers. • Z = {· · · , −1, 0, 1, 2, ···}, the set of integers. • R, the set of real numbers. • R+, the set of non-negative real numbers. Edward A. Lee | UC Berkeley — EECS3/32 Partial Orders — Basics Relations and Functions • A binary relation from A to B is a subset of A × B. • A partial function f from A to B is a relation where (a, b) ∈ f and (a, b0) ∈ f =⇒ b = b0. Such a partial function is written f : A*B. • A total function or just function f from A to B is a partial function where for all a ∈ A, there is a b ∈ B such that (a, b) ∈ f. Edward A. Lee | UC Berkeley — EECS4/32 Partial Orders — Basics Notation • A binary relation: R ⊆ A × B. • Infix notation: (a, b) ∈ R is written aRb. • A symbol for a relation: • ≤⊂ N × N • (a, b) ∈≤ is written a ≤ b. • A function is written f : A → B, and the A is called its domain and the B its codomain.
    [Show full text]
  • A General Account of Coinduction Up-To Filippo Bonchi, Daniela Petrişan, Damien Pous, Jurriaan Rot
    A General Account of Coinduction Up-To Filippo Bonchi, Daniela Petrişan, Damien Pous, Jurriaan Rot To cite this version: Filippo Bonchi, Daniela Petrişan, Damien Pous, Jurriaan Rot. A General Account of Coinduction Up-To. Acta Informatica, Springer Verlag, 2016, 10.1007/s00236-016-0271-4. hal-01442724 HAL Id: hal-01442724 https://hal.archives-ouvertes.fr/hal-01442724 Submitted on 20 Jan 2017 HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. A General Account of Coinduction Up-To ∗ Filippo Bonchi Daniela Petrişan Damien Pous Jurriaan Rot May 2016 Abstract Bisimulation up-to enhances the coinductive proof method for bisimilarity, providing efficient proof techniques for checking properties of different kinds of systems. We prove the soundness of such techniques in a fibrational setting, building on the seminal work of Hermida and Jacobs. This allows us to systematically obtain up-to techniques not only for bisimilarity but for a large class of coinductive predicates modeled as coalgebras. The fact that bisimulations up to context can be safely used in any language specified by GSOS rules can also be seen as an instance of our framework, using the well-known observation by Turi and Plotkin that such languages form bialgebras.
    [Show full text]
  • Hopf Monoids of Ordered Simplicial Complexes
    HOPF MONOIDS OF ORDERED SIMPLICIAL COMPLEXES FEDERICO CASTILLO, JEREMY L. MARTIN, AND JOSE´ A. SAMPER ABSTRACT. We study pure ordered simplicial complexes (i.e., simplicial complexes with a linear order on their ground sets) from the Hopf-theoretic point of view. We define a Hopf class to be a family of pure ordered simplicial complexes that give rise to a Hopf monoid under join and deletion/contraction. The prototypical Hopf class is the family of ordered matroids. The idea of a Hopf class allows us to give a systematic study of simpli- cial complexes related to matroids, including shifted complexes, broken-circuit complexes, and unbounded matroids (which arise from unbounded generalized permutohedra with 0/1 coordinates). We compute the antipodes in two cases: facet-initial complexes (a much larger class than shifted complexes) and unbounded ordered matroids. In the latter case, we embed the Hopf monoid of ordered matroids into the Hopf monoid of ordered generalized permuto- hedra, enabling us to compute the antipode using the topological method of Aguiar and Ardila. The calculation is complicated by the appearance of certain auxiliary simplicial complexes that we call Scrope complexes, whose Euler characteristics control certain coeffi- cients of the antipode. The resulting antipode formula is multiplicity-free and cancellation- free. CONTENTS 1. Introduction2 1.1. Background: Matroids and combinatorial Hopf theory2 1.2. Ordered simplicial complexes3 1.3. Polyhedra5 1.4. Antipodes6 2. Background and notation7 2.1. Simplicial complexes8 2.2. Matroids9 2.3. Set compositions, preposets, and the braid fan9 2.4. Generalized permutohedra 11 2.5. Species and Hopf monoids 14 2.6.
    [Show full text]
  • Lattice Duality: the Origin of Probability and Entropy
    , 1 Lattice Duality: The Origin of Probability and Entropy Kevin H. Knuth NASA Ames Research Center, Mail Stop 269-3, Moffett Field CA 94035-1000, USA Abstract Bayesian probability theory is an inference calculus, which originates from a gen- eralization of inclusion on the Boolean lattice of logical assertions to a degree of inclusion represented by a real number. Dual to this lattice is the distributive lat- tice of questions constructed from the ordered set of down-sets of assertions, which forms the foundation of the calculus of inquiry-a generalization of information theory. In this paper we introduce this novel perspective on these spaces in which machine learning is performed and discuss the relationship between these results and several proposed generalizations of information theory in the literature. Key words: probability, entropy, lattice, information theory, Bayesian inference, inquiry PACS: 1 Introduction It has been known for some time that probability theory can be derived as a generalization of Boolean implication to degrees of implication repre- sented by real numbers [11,12]. Straightforward consistency requirements dic- tate the form of the sum and product rules of probability, and Bayes’ thee rem [11,12,47,46,20,34],which forms the basis of the inferential calculus, also known as inductive inference. However, in machine learning applications it is often times more useful to rely on information theory [45] in the design of an algorithm. On the surface, the connection between information theory and probability theory seems clear-information depends on entropy and entropy Email address: kevin.h. [email protected] (Kevin H.
    [Show full text]
  • Hyperbolic Disk Embeddings for Directed Acyclic Graphs
    Hyperbolic Disk Embeddings for Directed Acyclic Graphs Ryota Suzuki 1 Ryusuke Takahama 1 Shun Onoda 1 Abstract embeddings for model initialization leads to improvements Obtaining continuous representations of struc- in various of tasks (Kim, 2014). In these methods, sym- tural data such as directed acyclic graphs (DAGs) bolic objects are embedded in low-dimensional Euclidean has gained attention in machine learning and arti- vector spaces such that the symmetric distances between ficial intelligence. However, embedding complex semantically similar concepts are small. DAGs in which both ancestors and descendants In this paper, we focus on modeling asymmetric transitive of nodes are exponentially increasing is difficult. relations and directed acyclic graphs (DAGs). Hierarchies Tackling in this problem, we develop Disk are well-studied asymmetric transitive relations that can be Embeddings, which is a framework for embed- expressed using a tree structure. The tree structure has ding DAGs into quasi-metric spaces. Existing a single root node, and the number of children increases state-of-the-art methods, Order Embeddings and exponentially according to the distance from the root. Hyperbolic Entailment Cones, are instances of Asymmetric relations that do not exhibit such properties Disk Embedding in Euclidean space and spheres cannot be expressed as a tree structure, but they can be respectively. Furthermore, we propose a novel represented as DAGs, which are used extensively to model method Hyperbolic Disk Embeddings to handle dependencies between objects and information flow. For exponential growth of relations. The results of example, in genealogy, a family tree can be interpreted as a our experiments show that our Disk Embedding DAG, where an edge represents a parent child relationship models outperform existing methods especially and a node denotes a family member (Kirkpatrick, 2011).
    [Show full text]